/prediction-src

Primary LanguagePython

ReadMe

Steps 1: Clone This Repo :

    git clone https://github.com/someshjaishwal/prediction-src.git

Step 2: Create virtual environment:

    cd prediction-src
    virtualenv -p python3 venv
    source venv/bin/activate

Step 3: Install dependencies:

    cat requirements.txt | xargs -n 1 pip install

Predict transcript of a single wav file

To predict transcription of an audio named audio.wav, use below command from prediction-src/ directory:

CUDA_VISIBLE_DEVICE=0 python predict.py --cuda --model-path models/network.pth --audio-path audio.wav

Evaluate the pretrained model using manifest.csv

You can also use a manifest.csv file to evaluate the pretrained model. Each row of the manifest.csv file contains path of an audio and transcript of that audio, separated by comma(,) delimiter. e.g. :

../data/speaker1/audio1.wav,chilli flakes
../data/speaker1/audio2.wav,mint mayonnaise

NOTE: Every character of the transcription is small. If you're using manifest.csv, make sure spelling of each transcription in manifest.csv matches with files/transcriptions.py

Finally, to evaluate the pretrained model using manifest.csv, hit the below command from prediction-src/ directory

CUDA_VISIBLE_DEVICES=0 python evaluate.py --cuda --model-path models/network.pth --manifest manifest.csv

Needless to mention, at any time we choose CUDA_VISIBLE_DEVICES of our choice